Re: RFC on PCI Device Lock

[Alex Williamson <alex.williamson@redhat.com>]

On Wed, 5 Sep 2018 07:26:52 -0700
Sinan Kaya <okaya@kernel.org> wrote:

> On 9/4/2018 9:13 PM, Alex Williamson wrote:
> >> Let me know if I misunderstood you. I'll again wait for test feedback first.  
> > Yep, that's essentially it.  I assume we'd want to trickle this through
> > other interfaces and the hfi1 folks can look at whether they want to
> > drop their custom config save/restore in favor of using
> > pci_reset_function_locked(), which would do that for them.
> > pci_reset_bus() should also allow the caller to set a mask so vfio
> > doesn't worry about defaults changing in the core.  Thanks,  
> The testing was successful.
> 
> "--- Comment #15 from Dennis (dennis.dalessandro@intel.com) ---
> I have tested the two new patches applied on top of 4.19-rc1 and they seem to
> be working as well."
> 
> Before I post the code on the list, I want to make sure I capture your
> request correctly.
> 
> You are asking for more APIs to have the mask feature. I can add
> slot and bus reset options to pci_reset_bus().
> 
> If there are other functions that need mask feature, we can deal
> with them later as I want to focus on the regression short term.
> 
> Though, it would be good to enumerate the functions with needed masks
> to start the conversation.

Well, if __pci_reset_function_locked() has this option, would only make
sense that these would as well:

pci_probe_reset_function()
pci_reset_function()
pci_reset_function_locked()
pci_try_reset_function()

Then I'd like to see the same sort of caller specification added to
pci_reset_bus() so that vfio can specify whether it's expecting a sbr
or slot reset.  Thanks,

Alex